A time varying speed of light as a solution to cosmological puzzles

Andreas Albrecht and João Magueijo

Theoretical Physics, The Blackett Laboratory, Imperial College, Prince Consort Road, London, SW7 2BZ, U.K.
We consider the cosmological implications of light travelling faster in the early Universe. We propose
a prescription for deriving corrections to the cosmological evolution equations while the speed of
light c is changing. We then show how the horizon, flatness, and cosmological constant problems
may be solved. We also study cosmological perturbations in this scenario and show how one may
solve the homogeneity and isotropy problems. As it stands, our scenario appears to most easily
produce extreme homogeneity, requiring structure to be produced in the Standard Big Bang epoch.
Producing significant perturbations during the earlier epoch would require a rather careful design
of the function c(t). The large entropy inside the horizon nowadays can also be accounted for in
this scenario.
arXiv:astro-ph/9811018v2 5 Jan 1999

PACS Numbers: 98.80.Cq, 98.80.-k, 95.30.Sf

I. THE PUZZLES OF THE BIG BANG MODEL
Cosmologists have long been dissatisfied with the
“Standard Big Bang” (SBB) model of the Universe. This
is not due to any conflict between the big bang theory
and observations, but because of the limited scope of-
fered by the SBB to explain certain striking features of
the Universe. From the SBB perspective the homogene-
ity, isotropy, and “flatness” of the Universe, and the pri-
mordial seeds of galaxies and other structure are all fea-
tures which are “built in” from the beginning as initial
conditions. Cosmologists would like to explain these fea-
tures as being the result of calculable physical processes.
A great attraction of the Inflationary Cosmologies [1] is
that they address these issues by showing on the basis of
concrete calculations that a wide variety of initial condi-
tions evolve, during a period of cosmic inflation, to re-
flect the homogeneity, isotropy, flatness and perturbation
spectrum that we observe today.
So far, all attempts to achieve this kind of improve-
ment over the SBB have wound up taking the basic infla-
tionary form, where the observable Universe experiences
a period of “superluminal” expansion. This is accom-
plished by modifying the matter content of the Universe
in such a way that ordinary Einstein gravity becomes
repulsive and drives inflationary expansion. This pro-
cess is in many ways remarkably straightforward and has
found numerous realizations over the years ( [2–5], etc),
although it might still be argued that a truly compelling
microscopic foundation for inflation has yet to emerge.
One interesting question is whether inflation is the
right solution to the cosmological puzzles. Is inflation
really what nature has chosen to do? When this mat-
ter is discussed there is a notable absence of any real
competition to inflation, and this must be counted in in-
flation’s favour. However, we believe the picture would
become much clearer if some kind of debate along these
lines were possible. To this end, we discuss here a possi-
ble alternative to inflationary cosmology which, while not
as well developed as today’s inflationary models, might
lead to some illuminating discussion.
In this alternative picture, rather than changing the
matter content of the Universe, we change the speed of
light in the early Universe. We assume that the Uni-
verse matter content is the same as in the SBB, that is,
the Universe is radiation dominated at early times. We
also assume that Einstein’s gravity is left unchanged, in
a sense made precise in Section IV. The geometry and
expansion factor of the Universe are therefore the same
as in the SBB. However the local speed of light, as mea-
sured by free falling observers associated with the cosmic
expansion, varies in time, decelerating from a very large
value to its current value.
We discuss below how Varying Speed of Light (VSL)
models might resolve the same cosmological puzzles as
inflation, and offer a resolution to the cosmological con-
stant problem as well. We shall not dwell on the possible
mechanisms by means of which the speed of light could
have changed. Rather we wish to concentrate on the
conditions one should impose on VSL models for their
cosmological implications to be interesting. This phe-
nomenological approach should be regarded as a curios-
ity, which, we hope, will prompt further work towards an
actual theory in which the physical basis of VSL models
is realized.
One may doubt that such a self-consistent theory could
ever be constructed. We therefore feel forced to tran-
scend the scope of this paper, and discuss essential as-
pects of such a theory. We find it befitting to start our
discussion with an assessment of the experimental mean-
ing of a varying c (Section II). We also need to be more
specific about VSL theories in order to tackle the flatness,
cosmological constant, homogeneity, and entropy prob-
lems. In Section IV we state what is actually required
from any VSL theory to solve these problems. However in
Appendix I we lay out the foundations for such a theory.
II. THE MEANING OF A VARIABLE SPEED OF
LIGHT
We first address the question of the meaning of a vary-
ing speed of light. Could such a phenomenon be proved
or disproved by experiment? Physically it does not make

1
sense to talk about constancy or variability of any di-
mensional “constant”. A measurement of a dimensional
quantity must always represent its ratio to some stan-
dard unit. For example, the length of my arm in meters
is really the dimensionless quantity given by the ratio of
the arm length to the length of a meter stick. If the ratio
varied, one could interpret this as a variation in either
(or both) of the two lengths. In familiar situations, there
is usually a preferred interpretation which distinguishes
itself by giving a simpler view of the world. Choosing a
given person’s arm as a standard of length would require
a whole range of simple objects to undergo peculiar dy-
namics, whereas assuming the meter stick to be constant
would usually give a much simpler picture.
None the less, a given theory of the world requires di-
mensional parameters. If these parameters varied, how
would that process show up in experiments? Suppose we
set out to measure the speed of light. For this one needs
a length measure (rod) and a clock. In a world described
by a theory with time varying dimensional parameters,
it is quite possible that the rods and clocks, as well as
the photon speeds, could all vary. Because measurements
are fundamentally dimensionless, the experimental result
will only measure some dimensionless combination of the
fundamental constants. Let us sketch a simple illustra-
tion: Suppose we measure time with an atomic clock.
Taking the Rydberg energy (ER = me e4 /2(4πǫo )2 h̄2 ) to
represent the dependence of all atomic energy levels on
the fundamental constants, the oscillation period of the
atomic clock will be ∝ h̄/ER . Likewise, taking the Bohr
radius (a0 = 4πǫ0 h̄2 /me e2 ) to reflect the relationship be-
tween the lengths of ordinary objects (made of atoms)
and the fundamental constants, the length of our rod is
∝ a0 . Thus a measurement of c with our equipment is
really a measurement of the dimensionless quantity
c 8πǫ0
= (1)
a0 /(h̄/ER ) α
essentially the fine structure constant. We could of course
use other equipment which depends in different ways on
the fundamental dimensionless constants. For example,
pendulum clocks will necessarily involve Newton’s con-
stant G. Different experiments will result, which measure
different dimensionless combinations of the fundamental
dimensional constants. Our conclusion that physical ex-
periments are only sensitive to dimensionless combina-
tions of dimensional constants is hardly a new one. This
idea has been often stressed by Dicke (eg. [21]), and we
believe this is not controversial.
Thus, speaking in theoretical terms of time varying di-
mensional constants can lead to problems. To give an
historical example, papers [18,19] were written claiming
stringent experimental upper bounds on the time vari-
ability of the dimensional quantity h̄c. In these the prod-
uct Eλ was found to be the same for light emitted at very
different redshifts. From the deBroglie relation h̄c = Eλ
one infers the constancy of h̄c. Bekenstein gives an illumi-
nating discussion of the fallacy built into this argument
2
[20]. Built into E ∝ 1/a and λ ∝ a is the assump-
tion that h̄c is constant, for otherwise the wavevector k µ
and the momentum vector pµ could not both be parallel
transported. Hence the experimental statement that h̄c
is constant is circular.
What would we do therefore if we were to observe
changing dimensionless quantities? Any theory explain-
ing the phenomenon would necessarily have to make use
of dimensional quantities. It would a priori be a matter of
choice, prejudice, or convenience to decide which dimen-
sional quantities are variable and which are constant (as
we mentioned in the illustration above). There would be
a kind of equivalence, or duality between theories based
on any two choices as far as dimensionless observations
are concerned. However, the equations for two theories
which are observationally equivalent, but which have dif-
ferent dimensional parameters varying, will in general not
look the same, and again simplicity will end up being an
important factor in making a choice between theories. In
what follows, we will prefer to work with models which
have the simplicity of “minimal coupling”.
Let us illustrate this point with a topical example.
There has been a recent claim [22] of experimental ev-
idence for a time changing fine structure constant α =
e2 /(4πh̄c). Although the ongoing chase for systematics
precludes any definitive conclusions, let us assume for the
purpose of the argument that the effect is real.
In building a theory which explains a variable α we
must make a decision. We could postulate that electric
charge changes in time, or, say, that h̄c must change in
time. Bekenstein [23] constructs a theory based on the
first alternative. He postulates a Lorentz invariant ac-
tion, which does not conserve electric charge. Our the-
ory is based on the second choice. We postulate breaking
Lorentz invariance, a changing h̄c, and consequently non-
conservation of energy. Any arguments against the ex-
perimental meaning of a changing c can also be directed
at Bekensteins’ changing e theory, and such arguments
are in both cases meaningless. In both cases the choice of
a changing dimensional “constant” reverts to the postu-
lates of the theory and is not, a priori, an experimental is-
sue. The observables are always dimensionless. However,
the minimally coupled theories based on either choice are
not dual (as we shall point out in Appendix I). For this
reason one might prefer one formulation over the other.
Finally, and on a different tone, suppose that future
experiments were to confirm that not only α changes in
time, but also that there are time variations in dimen-
sionless coupling constants based on other interactions,
αi = gi2 /(h̄c)1 . Suppose further that the ratios between
the various constants, rij = αi /αj , were observed to be
1
In writing these constants we have assumed that the cou-
plings of these interactions are defined in terms of “charges”
(with dimensions of [E]1/2 [L]1/2 ).
 TIME
US NOW
TWO REGIONS IN OUR PAST
WHICH ARE CAUSALLY PAST LIGHT CONE
DISCONNECTED
SOME TIME IN OUR PAST
BIG BANG
SPACE
FIG. 1. Conformal diagram (light at 45 ) showing the hori-
◦
zon structure in the SBB model. Our past light cone contains
regions outside each others’ horizon.
constant. Choosing what dimensional constants were in-
deed constants would still be a matter of taste. One could
still define a theory in which the various charges gi change
in time, with fixed ratios, and h̄c remains constant. How-
ever it would perhaps start to make more sense, merely
for reasons of simplicity, to postulate instead a changing
h̄c.
Therefore, even though a variable c cannot be made a
dimensionless statement, evidence in favour of theoretical
models with varying c could be accrued if the other αi
changed, with fixed ratios.
III. COSMOLOGICAL HORIZONS
Perhaps the most puzzling feature of the SBB is the
presence of cosmological horizons. At any given time
any observer can only see a finite region of the Universe,
with comoving radius rh = cη, where η denotes confor-
mal time, and c the speed of light. Since the horizon size
increases with time we can now observe many regions in
our past light cone which are causally disconnected, that
is, outside each others’ horizon (see Fig. 1). The fact
that these regions have the same properties (eg. Cosmic
Microwave background temperatures equal to a few parts
in 105 ) is puzzling as they have not been in physical con-
tact. This is a mystery one may simply relegate to the
setting up of initial conditions in our Universe.
One may however try to explain these very peculiar
initial conditions. The horizon problem is solved by in-
flationary scenarios by postulating a period of accelerated
or superluminal expansion, that is, if a is the expansion
factor of the Universe, a period with ä > 0. The Fried-
man equations require that the strong energy condition
ρ + 3p/c2 ≥ 0 must then be violated, where ρc2 and p
are the energy density and pressure of the cosmic matter.
This violation is achieved by the inflaton field. If ä > 0
for a sufficiently long period one can show that cosmo-
logical horizons are a post-inflation illusion, and that the
whole observed Universe has in fact been in causal con-
tact since an early time.
A more minimalistic way of solving this problem is to
postulate that light travelled faster in the Early Universe.
3
TIME
US NOW
t
HORIZON AT c
PAST LIGHT CONE
tc
BIG BANG
SPACE
FIG. 2. Diagram showing the horizon structure in a SBB
model in which at time tc the speed of light changed from c−
to c+ ≪ c− . Light travels at 45◦ after tc but it travels at
a much smaller angle with the space axis before tc . Hence
it is possible for the horizon at tc to be much larger than
the portion of the Universe at tc intersecting our past light
cone. All regions in our past have then always been in causal
contact.
Suppose there was a “phase transition” at time tc when
the speed of light changed from c− to c+ . Our past light
cone intersects t = tc at a sphere with comoving radius
r = c+ (η0 − ηc ), where η0 and ηc are the conformal times
now and at tc . This is as much of the Universe after the
phase transition as we can see today [6]. On the other
hand the horizon size at tc has comoving radius rh =
c− ηc . If c− /c+ ≫ η0 /ηc , then r ≪ rh , meaning that the
whole observable Universe today has in fact always been
in causal contact (see Fig. 2). Some simple manipulations
show that this requires
c− 1 1 T+
log10 +
≫ 32 − log10 zeq + log10 c+ (2)
c 2 2 TP
where zeq is the redshift at matter radiation equality, and
Tc+ and TP+ are the Universe and the Planck tempera-
tures after the phase transition. If Tc+ ≈ TP+ this implies
light travelling more than 30 orders of magnitude faster
before the phase transition. It is tempting, for symmetry
reasons, simply to postulate that c− = ∞ but this is not
strictly necessary.
IV. A PRESCRIPTION FOR MODIFYING
PHYSICAL LAWS WHILE THE SPEED OF
LIGHT IS VARYING
Hidden in the above argument is the assumption that
the geometry of the Universe is not affected by a chang-
ing c. We have allowed a changing c to do the job nor-
mally done by “superluminal expansion”. To enhance
this effect we have forced the geometry to still be the
SBB geometry. We now elaborate on this assumption.
We will propose a prescription for how, in general, to
modify gravitational laws while c is changing. This pre-
scription is merely the one we found the most fertile. In
Appendix I we describe in detail a theory which realizes
this prescription.
 The basic assumption is that a variable c does not in- proportional to c2 . If, however, K 6= 0 not even mass is
duce corrections to curvature in the cosmological frame, conserved.
and that Einstein’s equations, relating curvature to stress In Eqn. 5 we have included the effects of Ġ under the
energy, are still valid. The rationale behind this postulate same postulate merely for completness. In such a formu-
is that c changes in the local Lorentzian frames associ- lation VSL does not reduce to Brans Dicke theory when
ated with cosmological expansion. The effect is a special ċ = 0, and Ġ 6= 0. This is because we postulate that
relativistic effect, not a gravitational effect. Therefore Friedmann equations remain unchanged, which implies
curvature should not feel a changing c. that the conservation equations acquire terms in ċ and
The previous statement is not covariant. However in- Ġ. In Brans Dicke theory one postulates exactly the op-
troducing a function c(t) is not even Lorentz invariant. posite: the conservation equations must still be valid, so
So it is not surprising that a favoured gauge, or coor- that the weak equivalence principle is satisfied. While we
dinate choice, must be made, where the function c(t) is could have taken this stance for c as well we feel that vio-
specified, and in which the above postulate holds true. lation of energy conservation is the hallmark of changing
The cosmological frame (with the cosmological time t) c. Variable c must break Poincare invariance, for which
provides such a preferred frame. energy is the Noether current. Barrow [25] has proposed
In a cosmological setting the postulate proposed im- a formulation of VSL which has the correct Brans Dicke
plies that Friedman equations remain valid even when limit.
ċ 6= 0:
 2
ȧ 8πG Kc2 V. THE FLATNESS PUZZLE
= ρ− 2 (3)
a 3 a
ä 4πG  p We now turn to the flatness puzzle. The flatness puzzle
=− ρ+3 2 (4)
a 3 c can be illustrated as follows. Let ρc be the critical density
of the Universe:
where, we recall, ρc2 and p are the energy and pressure
densities, K = 0, ±1 and G the curvature and the gravi- 3
 2
ȧ
tational constants, and the dot denotes a derivative with ρc = (6)
8πG a
respect to proper time. If the Universe is radiation dom-
inated, p = ρc2 /3, and we have as usual a ∝ t1/2 . We that is, the mass density corresponding to K = 0 for a
have assumed that a frame exists where c = c(t), and given value of ȧ/a. Let us define ǫ = Ω−1 with Ω = ρ/ρc .
identified this frame with the cosmological frame. Then
The assumption that Einstein’s equations remain unaf-  
fected by decelerating light carries with it an important ρ̇ ρ̇c
ǫ̇ = (1 + ǫ) − (7)
consequence. Bianchi identities apply to curvature, as ρ ρc
a geometrical identity. These then imply stress energy
conservation as an integrability condition for Einstein’s If p = wρc2 (with ẇ = 0), using Eqns.(3), (4), and (5)
equations. If ċ 6= 0, however, this integrability condition we have:
is not stress energy conservation. Source terms, propor-
tional to ċ/c, come about in the conservation equations. ρ̇ ȧ Ġ ċ ǫ
= −3 (1 + w) − + 2 (8)
Seen in another way, the conservation equations im- ρ a G c1+ǫ
ply an equation of motion for free falling point particles. ρ̇c ȧ Ġ
This is normally the geodesic equation, but now source = − (2 + (1 + ǫ)(1 + 3w)) − (9)
ρc a G
terms will appear in the geodesic equation. Clearly a vi-
olation of the weak equivalence principle is implied while and so
c is changing [7]. This, of course, does not conflict with
ȧ ċ
experiment, as we take ċ 6= 0 only in the Early Universe, ǫ̇ = (1 + ǫ)ǫ (1 + 3w) + 2 ǫ (10)
possibly for only a very short time (such as a phase tran- a c
sition). In the SBB ǫ grows like a2 in the radiation era, and like a
Although this is a general remark we shall be concerned in the matter era, leading to a total growth by 32 orders
mostly with violations of energy conservation in a cosmo- of magnitude since the Planck epoch. The observational
logical setting. Friedman equations can be combined into fact that ǫ can at most be of order 1 nowadays requires
a “conservation equation” with source terms in ċ/c and that either ǫ = 0 strictly, or an amazing fine tuning must
Ġ/G: have existed in the initial conditions (ǫ < 10−32 at t =
tP ). This is the flatness puzzle.
ȧ  p Ġ 3Kc2 ċ
ρ̇ + 3 ρ + 2 = −ρ + (5) The ǫ = 0 solution is in fact unstable for any matter
a c G 4πGa2 c field satisfying the strong energy condition 1 + 3w > 0.
In a flat Universe (K = 0) a changing c does not vio- Inflation solves the flatness problem with an inflaton field
late mass conservation. Energy, on the other hand, is which satisfies 1 + 3w < 0. For such a field ǫ is driven

4
  4
√

c (R + 2Λ1 )
Z
towards zero instead of away from it. Thus inflation can S= dx4 −g + LM + LΛ2 (11)
solve the flatness puzzle. 16πG
As Eqn. 10 shows a decreasing speed of light (ċ/c < 0)
would also drive ǫ to 0. If the speed of light changes in a where LM is the matter fields Lagragian. The term in Λ1
sharp phase transition, with |ċ/c| ≫ ȧ/a, we can neglect is a geometrical cosmological constant, as first introduced
the expansion terms in Eqn. 10. Then ǫ̇/ǫ = 2ċ/c so that by Einstein. The term in Λ2 represents the vacuum en-
ǫ ∝ c2 . A short calculation shows that the condition (2) ergy density of the quantum fields [11]. Both tend to
also ensures that ǫ ≪ 1 nowadays, if ǫ ≈ 1 before the dominate the energy density of the Universe, leading to
transition. the so-called cosmological constant problem. However
The instability of the K 6= 0 Universes while ċ/c < 0 they represent two rather different problems. We shall
can be expected simply from inspection of the non con- attempt to solve the problem associated with the first,
servation equation Eq. (5). Indeed if ρ is above its criti- not the second, term. Ususally one hopes that the sec-
cal value, then K = 1, and Eq. (5) tells us that mass is ond term will be cancelled by an additional couter-term
taken out of the Universe. If ρ < ρc , then K = −1, and in the Lagrangian. In the rest of this paper it is the
then mass is produced. Either way the mass density is geometrical cosmological constant that is under scrutiny.
pushed towards its critical value ρc . In contrast with the If the cosmological constant Λ 6= 0 then the argument
Big Bang model, during a period with ċ/c < 0 only the in the previous section still applies, with ρ = ρm + ρΛ ,
K = 0 Universe is stable. where ρm is the mass density in normal matter, and
Note that with the set of assumptions we have used a Λc2
changing G cannot solve the flatness problem (cf. [8–10]). ρΛ = (12)
We have assumed in the previous discussion that we 8πG
are close, but not fine-tuned, to flatness before the tran- is the mass density in the cosmological constant. One
sition. It is curious to note that this need not be the still predicts Ωm + ΩΛ = 1, with Ωm = ρm /ρc and ΩΛ =
case. Suppose instead that the Universe acquires “natu- ρΛ /ρc . However now we also have
ral initial conditions” (eg. ǫ ≈ 1) well before the phase
transition occurs. If such Universes are closed they rec- ȧ  pm  Ġ 3Kc2 ċ
ρ̇m + 3 ρm + 2 = −ρ̇Λ − ρ + (13)
ollapse before the transition. If they are open, then they a c G 4πGa2 c
approach ǫ = −1. This is the Milne Universe, which is
our case (constant G) may be seen as Minkowski space- If Λ is indeed a constant then from Eq. (12)
time. Such a curvature dominated Universe is essentially
ρ̇Λ ċ Ġ
empty, and a coordinate transformation can transform it =2 − (14)
into Minkowski space-time. Inflation cannot save these ρΛ c G
empty Universes, as can be seen from Eqn. 10. Indeed If we define ǫΛ = ρΛ /ρm we then find, after some straight-
even if 1 + 3w < 0 the first term will be negligible if forward algebra, that
ǫ ≈ −1. This is not true for VSL: the second term will
still push an ǫ = −1 Universe towards ǫ = 0.
 
ȧ ċ 1 + ǫΛ
Heuristically this results from the fact that the viola- ǫ̇Λ = ǫΛ 3 (1 + w) + 2 (15)
a c 1+ǫ
tions of energy conservation responsible for pushing the
Universe towards flatness do not depend on there being Thus, in the SBB model, ǫΛ increases like a4 in the ra-
any matter in the Universe. This can be seen from in- diation era, like a3 in the matter era, leading to a total
spection of Eqn. (5). growth by 64 orders of magnitude since the Planck epoch.
In this type of scenario it does not matter how far be- Again it is puzzling that ǫΛ is observationally known to
fore the transition the “initial conditions” are imposed. be at most of order 1 nowadays. We have to face another
We end up with a chaotic scenario in which Darwinian fine tuning problem in the SBB model: the cosmological
selection gets rid of all the closed Universes. The open constant problem.
Universes become empty and cold. In the winter of these If ċ = 0 the solution ǫΛ = 0 is in fact unstable for any
Universes a phase transition in c occurs, producing mat- w > −1. Hence violating the strong energy condition
ter, and leaving the Universe very fine tuned, indeed as 1 + 3w > 0 would not solve this problem. Even in the
an Einstein deSitter Universe (EDSU). limiting case w = −1 the solution ǫΛ = 0 is not an attrac-
tor: ǫΛ would merely remain constant during inflation,
then starting to grow like a4 after inflation. Therefore
VI. THE COSMOLOGICAL CONSTANT inflation cannot “explain” the small value of ǫΛ , as it can
PROBLEM with ǫ, unless one violates the dominant energy condition
w ≥ −1.
There are two types of cosmological constant problems, However, as Eqn. (15) shows, a period with ċ/c ≪ 0
and we wish to start our discussion by differentiating would drive ǫΛ to zero. If the speed of light changes
them. Let us write the action as: suddenly (|ċ/c| ≫ ȧ/a) then we can neglect terms in
ȧ/a, and so

5
 ǫ̇Λ ċ 1 many regions which have apparently been causally dis-
=2 (16)
ǫΛ (1 + ǫΛ ) c1+ǫ connected. Although solving the horizon problem is a
necessary condition for solving the homogeneity problem,
which when combined with ǫ̇/ǫ = 2ċ/c leads to in a generic inflationary model solving the first causes se-
ǫΛ ǫ rious problems in solving the latter. Early causal contact
∝ (17) between the entire observed Universe allows equilibration
1 + ǫΛ 1+ǫ
processes to homogenize the whole observed Universe. It
The exact constraint on the required change in c depends is crucial to the inflation picture that before inflation the
on the initial conditions in ǫ and ǫΛ . In any case once observable universe in well inside the Jeans length, and
both ǫ ≈ 1 and ǫΛ ≈ 1 we have ǫΛ ∝ c2 . Then we thus equilibrates toward a homogeneous state. However
can solve the cosmological constant problem in a sudden no such process is perfect, and small density fluctuations
phase transition in which tend to be left outside the Hubble radius, once the Uni-
verse resumes its standard Big Bang course. These fluc-
c− 1 Tc+ tuations then grow like a2 during the radiation era, like
log10 ≫ 64 − log 10 z eq + 2 log 10 (18)
c+ 2 TP+ a during the matter era, usually entailing a very inhomo-
geneous Universe nowadays. This is a common flaw in
This condition is considerably more restrictive than (2), early inflationary models [12] which requires additional
and means a change in c by more than 60 orders of mag- fine-tuning to resolve.
nitude, if Tc+ ≈ TP+ . Note that once again a period with In order to approach this problem we study in Ap-
Ġ/G would not solve the cosmological constant problem. pendix II the effects of a changing c on the theory of
Equations (10) and (15) are the equations one should scalar cosmological perturbations [13]. The basic re-
integrate to find conditions for solving the flatness and sult is that the comoving density contrast ∆ and gauge-
cosmological constant problems for arbitrary initial con- invariant velocity v are subject to the equations:
ditions and with arbitrary curves c(t). They generalize  
the conditions (2) and (18) which are valid only for a ′ a′ c′ a′
∆ − 3w + ∆ = −(1 + w)kv − 2 wΠT (19)
starting point with ǫ ≈ 1 and ǫΛ ≈ 1 and for a step a c a
function c(t).  ′
a c′
 
c2s k

3 a′ a′ c′

As in the case of the flatness problem we do not need v′ + −2 v= − + ∆
to impose “natural initial conditions” (ǫΛ ≈ 1) just be- a c 1 + w 2k a a c
 ′ 2 !
fore the transition. These could have existed any time kc2 w 2/3 3 a
before the transition, and the argument would still go + Γ − kc + wΠT (20)
1+w 1 + w k 2 c2 a
through, albeit with a rather different overall picture for
the history of the Universe. where k is the wave vector of the fluctuations, and Γ is
If ǫΛ ≈ 1 well before the transition, then the Uni- the entropy production rate, ΠT the anisotropic stress,
verse soon becomes dominated by the cosmological con- and cs the speed of sound, according to definitions spelled
stant. We have inflation! The curvature and matter will out in Appendix II.
be inflated away. We end up in a de-Sitter Universe. In the case of a sudden phase transition Eqn. (19)
When the transition is about to occur it finds a flat Uni- shows us that ∆ ∝ c, regardless of the chosen equations
verse (ǫ = 0), with no matter (ρm = 0), and with a of state for Γ and ΠT . Hence
cosmological constant. If we rewrite Eqn.(15) in terms
of ǫm = ρm /ρΛ , for ǫ = 0 and |ċ/c| ≫ ȧ/a, we have ∆+ c+
ǫ̇m = −2(ċ/c)(1 + ǫm). Integrating leads to 1 + ǫm ∝ c−2 . = − (21)
∆− c
We conclude that we do not need the presence of any
matter in the Universe for a VSL transition to convert a meaning a suppression of any fluctuations before the
cosmological constant dominated Universe into a EDSU phase transition by more than a factor of 10−60 if con-
Universe full of ordinary matter. This can be seen from dition (18) is satisfied. The suppression of fluctuations
Eqns. (13)-(14). A sharp decline in c will always dis- induced by a sudden phase transition in c can be intu-
charge any vacuum energy density into ordinary matter. itively understood in the same fashion as the solution to
We stress the curious point that in this type of scenario the flatness problem. Mass conservation violation ensures
the flatness problem is not solved by VSL, but rather by that only a Universe at critical mass density is stable, if
the period of inflation preceding VSL. ċ/c ≪ 0. But this process occurs locally, so after the
phase transition the Universe should be left at critical
density locally. Hence the suppression of density fluctu-
VII. THE HOMOGENEITY OF THE UNIVERSE ations.
We next need to know what are the initial conditions
Solving the horizon problem by no means guarantees for ∆ and v. Suppose that at some very early time ti one
solving the homogeneity problem, that is, the uncanny has ċ/c = 0 and the whole observable Universe
√ nowadays
homogeneity of the currently observed Universe across is inside the Jeans length: η0 ≪ ci ηi / 3. The latter

6
condition is enforced as a byproduct of solving the hori-
zon problem. The whole observable Universe nowadays
is then initially in a thermal state. What is more each
portion of the Universe can be described by the canonical
ensemble and so the Universe is homogeneous apart from
thermal fluctuations [14]. These are characterized by the
mass fluctuation
2 hδM 2 i 4kb Ti
σM = = (22)
hM i2 M c2i
Converted into a power spectrum for ∆ this is a white
noise spectrum with amplitude
4kb Ti
P∆ (k) = h|∆(k)2 |i ∝ (23)
ρi c2i
What happens to a thermal distribution, its tempera-
ture, and its fluctuations, while c is changing? In ther-
mal equilibrium the distribution function of particle en-
ergies is the Planck distribution P (E) = 1/(eE/kb T − 1),
where T is the temperature. When one integrates over
the whole phase space, one obtains the bulk energy den-
sity ρc2 ∝ (kb T )4 /(h̄c)3 . Let us now consider the time
when the Universe has already flattened out sufficiently
for mass to be approximately conserved. To define the
situation more completely, we make two additional mi-
crophysical assumptions. Firstly, let mass be conserved
also for individual quantum particles, so that their en-
ergies scale like E ∝ c2 . Secondly, we assume particles’
wavelengths do not change with c. If homogeneity is pre-
served, indeed the wavelength is an adiabatic invariant,
fixed by a set of quantum numbers, eg: λ = L/n for a
particle in a box of size L.
Under the first of these assumptions a Planckian distri-
bution with temperature T remains Planckian, but T ∝
c2 . Under the second assumption, we have λ = 2πh̄c/E,
and so h̄/c should remain constant. Therefore the phase
space structure is changed so that, without particle pro-
duction, one still has ρc2 ∝ (kb T )4 /(h̄c)3 , with T ∝ c2 .
A black body therefore remains a black body, with a
temperature T ∝ c2 . If we combine this effect with ex-
pansion, with the aid of Eqn. (5) we have

ȧ ċ
 v ′ + (1 − 3w)
Ṫ + T −2 =0 (24)
a c
We can then integrate this equation through the epoch
when c is changing to find the temperature Ti of the
initial state. This fully fixes the initial conditions for
scalar fluctuations, by means of (23).
In the case of a sudden phase transition we have T + =
− 2+ 2−
T c /c , and so
but since ∆ ∝ c we have
s
+ 4kb T + c+
∆ (k) ≈ (27)
ρ+ c2+ c−
Even if T + = TP+ = 1019 Gev these fluctuations would
still be negligible nowadays. Therefore although the Uni-
verse ends up in a thermal state after the phase transi-
tion, its thermal fluctuations, associated with the canon-
ical ensemble, are strongly suppressed.
For a more general c(t) function the procedure is as
follows. Integrate Eqn. (24) backwards up to a time ti
when ċ = 0, to find T (ti ). Give ∆(ti ) a thermal spectrum
of fluctuations, according to (23), with T (ti ). With this
initial condition integrate Eqns.(19) and (20) (or even
better the second order equation (64) given in Appendix
II), to find ∆ nowadays.
It is conceivable that a careful design of c(t) would
leave fluctuations, once ċ = 0 again, with the right am-
plitude and spectrum to explain structure formation. In
particular c(t) may be designed so as to convert a white
noise spectrum into a scale-invariant spectrum. However
we feel that until a mechanism for inducing c(t) is found
such efforts are bound to look ludicrously contrived.
We feel that the power of VSL scenarios is precisely
in leaving the Universe very homogenous, after c has
stopped changing. This would then set the stage for
causal mechanisms of structure formation to do their job
[15,16].
VIII. THE ISOTROPY OF THE UNIVERSE
There is a sense in which there is an isotropy problem
in the SBB model, similar to the homogeneity problem.
We follow closely the remark made in [13], pp.26.
In Appendix III we write down the vector Einstein’s
equations in the vector gauge, and from them we derive
the vorticity “conservation” equation when ċ/c 6= 0. If v
is the vorticity (defined in Appendix) and ΠT the vector
stress, we have:
a′ c′ kc w
v−2 v =− ΠT (28)
a c 2 1+w
In the absence of driving stress, v remains constant dur-
ing the radiation dominated epoch, and decays like 1/a
in the matter epoch. In [13] it is further argued that the
relevant dimensionless quantity is
(k/a)v 1
ω= ∝ (1−9w)/2 (29)
(a′ /ca) a

2− 4kb T − 4kb T + Hence for w > 1/9 vorticity grows, leading to a further
σM = 2−
= (25) fine tuning problem.
Mc M c2+
This is most notably a problem if we accept the Planck
or
equipartition proposal, introduced in [17]. At Planck
4kb T + epoch there would then be a significant vorticity. De-
∆− (k)2 ≈ (26) pending on how one looks at it, this vorticity would then
ρ+ c2+

7
get frozen in or grow, leading to a very anisotropic Uni-
verse nowadays.
Whether or not this is a problem is clearly debatable.
In any case either inflation or VSL models could solve
this prospective problem. For w < −1/3 we have that v
decays faster than 1/a2 . Whatever dimensionless quan-
tity one chooses to look at, vorticity is therefore safely
inflated away. If ċ/c 6= 0 we have that v ∝ c2 . Again
any primordial vorticity is safely suppressed after a phase
transition in c satisfying any of the conditions (2) or (18).
IX. THE ENTROPY PROBLEM AND SETTING
THE INITIAL CONDITIONS
Let us first consider the SBB model. Let Sh be the en-
tropy inside the horizon, and σh = Sh /kB be its dimen-
sionless counterpart. σh is of order 1096 nowadays. If we
assume that the only scales in the cosmological model are
the ones provided by the fundamental constants, then at
tP the temperature is TP . At Planck time, σh (being
dimensionless) is naturally of order 1. In the SBB model
the horizon distance is dh = 2t in the radiation domi-
nated epoch, and ignoring mass thresholds t ∝ 1/T 2. If
evolution is adiabatic one then has (in a flat Universe)
 3
TP
σh (t) ≈ σh (tP ) . (30)
T us assume therefore that at t = t−P (the Planck time with
Since σh (tP ) ∼ 1, one has σh (t0 ) ∼ 1096 . Thus the large
entropy inside the horizon nowadays is a reflection of the
lack of scales beyond the ones provided by the funda-
mental constants, the fact that the horizon size is much
larger nowadays than at Planck time, and the flatness of
the Universe. One may rephrase the horizon and flatness
problems in terms of entropy [2]. However if one is willing
to accept the horizon structure and flatness of the Uni-
verse simply as features of the initial conditions (rather
than problems), there is no additional entropy problem.
There is a problem that arises if one tries to solve the
horizon problem, keeping the adiabatic assumption, by
means of superluminal expansion. This blows what at
Planck time is a region much smaller than the Planck
size into a comoving region containing the whole observ-
able Universe nowadays. This solves the horizon prob-
lem. However if evolution is adiabatic such a process im-
plies that σh (t0 ) ≪ 1. Stated in another way, since the
number of particles inside the horizon nh is of the same
order as σh , this implies an empty Universe nowadays.
More mathematically, if dh is the horizon proper dis-
tance, one has
σ̇h 3
= (31)
σh dh
Rt
where we have used dh = a dt′ /a. With any standard
2
matter (p > −ρc /3) the horizon grows like t. Accord-
ingly σh grows like a power of t. On the other hand the
8
horizon grows faster than t if p < −ρc2 /3: it grows ex-
ponentially if p = −ρc2 , and like tn (with n > 1) for
−ρc2 < p < −ρc2 /3. This provides the inflationary so-
lution to the horizon problem. However in the latter
case Eqn. (31) implies that σh decreases exponentially,
leading to σh (t0 ) ≪ 1. The way inflation bypasses this
problem is by dropping the adiabatic assumption. Indeed
during inflation the Universe supercools, and a period of
reheating follows the end of inflation2 .
In a VSL scenarios the detailed solution to the entropy
problem depends on when and what type of “natural con-
ditions” are given to the pre transition Universe. We first
derive equations for the entropy under varying c. From
s = (4/3)ρc2 /T , ρ ∝ T 4 /(h̄c)3 , and from Eqns 5 and 13
we obtain that the entropy of radiation satisfies
ṡ 3 ρ̇ ȧ 3 ċ ǫ(1 + ǫΛ ) 3 ċ
= = −3 + − ǫΛ (32)
s 4ρ a 2c 1+ǫ 2c
If the Universe is EDSU, there are no violations of mass
conservation, and entropy is conserved. However if the
Universe is open or has a positive cosmological constant,
then we have seen that there is creation of mass. Accord-
ingly there must be creation of particles, and entropy is
produced. If the Universe is closed, particles are taken
away, and the entropy decreases.
The most suspicious case is therefore if the Universe
was Einstein de-Sitter before the phase transition. Let
the constants before the transition) the entropy inside
the horizon (which has proper size c− t−P ) was of order 1.
Then the entropy inside the Hubble volume at t = t+ P,
before and after the transition, is
3  3
c + t+ a(t−

P)
σh (t+
P) = σh (t−
P)
P
≈1 (33)
c − t−
P a(t+
P)
where we have used t+ − − + 2
P /tP = (c /c ) . One takes a frac-
+ − 3
tion (c /c ) of the horizon volume before the transition
to make the Hubble volume after the transition. However
the entropy inside the horizon has increased since t+P by
the same factor. Therefore entropy conservation in this
case does not conflict with σh (t+
P ) ≈ 1 after the transi-
tion. One way of understanding this is that by imposing
flatness from the outset (before the transition) one has
already “solved” the entropy problem. Notice that the
above argument works for any value of tc /t+P.
Now consider the case where “natural” initial condi-
tions were also imposed at t−P , with Λ = 0. One should
have ǫ(t−P ) of order 1. We have already discussed how
the flatness problem is solved in this case, when large
empty curvature dominated universes are filled with a
2
This issue has been carefully analyzed in the context of
inflationary models and models with time varying G in [10]
(nearly perfectly) critical energy density during the tran- tion would not conflict with the dynamics of flatness and
+
sition. Open Universes become very empty, but they are Λ, as shown before, but now t− P = tP .
still pushed to EDSU at the transition. One may in-
tegrate (32) to find that s+ /s− = (1 + ǫ− )−3/4 . One
may also use Eqn. 10 to find that ǫ has evolved since X. CONCLUSIONS
− + + 2 − + 2
t− −
P to ǫ (tP ) + 1 ≈ (a(tP )/a(tP )) ≈ (tP /tP ) , where
we have used a ∝ t for the Milne Universe. Hence we We have shown how a time varying speed of light could
have that during the transition entropy is produced like provide a resolution to the well known cosmological puz-
s+ /s− = (t+ − 3/2
P /tP ) = (c− /c+ )3 . Given that a ∝ t for zles. These “VSL” models could provide an alternative
such Universes, the entropy before the transition in the to the standard Inflationary picture, and furthermore re-
proper volume of size c+ t+ P is solve the classical cosmological constant puzzle. At a
3  3 3 technical level, the proposed VSL picture is not nearly
c + t+ a(t− c+
 
P) as well developed as the inflationary one, and one pur-
S − (c+ t+
P) =
P
≈ (34)
c − t− a(t+ c− pose of this article is to stimulate further work on the
P P)
unresolved technical issues. We are not trying to take
that is there is practically no entropy in relevant volume an “anti-inflation” stand, but we do strongly feel that
before the transition. However we have that after the broadening the range of possible models of the very early
transition Universe would be very healthy for the field of cosmology,
 − 3 and would ultimately allow us to state in more concrete
c terms the extent to which one model is preferred.
σh (t+
P ) = S + + +
(c t P ) = S − + +
(c t P ) ≈1 (35)
c+ On a more fundamental level we hope to expand the
phenomenological approach presented in this paper into a
In such scenarios the Universe is rather cold and empty theory where the concept of (Poincare) symmetry break-
before the transition. However the transition itself re- ing provides the physical basis for VSL. Symmetry break-
heats the Universe. Notice that, like in the first case ing is also the central ingredient in causal theories of
discussed, the above argument works for any value of structure formation. We therefore hope to arrive at a
tc /t+
P. scenario where symmetry breaking provides a complete
If at t = t−
P one also has ǫΛ ≈ 1 then we have a scenario and consistent complement to the SBB model which can
in which the cosmological constant dominates, solves the resolve the standard puzzles as well as explain the origin
flatness problem, and is discharged into normal matter. of cosmic structure.
However if ρΛ ≈ ρ− −
P at t ≈ tP , then whatever the tran- Note added in proof:
sition time, after the transition the Universe will have a After this paper, as well as its sequel [26], were com-
density in normal matter equal to ρm = ρ− P . Hence the pleted J. Moffat brought to our attention two papers in
Hubble time after the transition will be t− P whatever the
, which he proposes a similar idea [27]. While Moffat’s
actual age of the Universe. One may integrate (32) to find work does not go as far as ours in addressing the flat-
that in this case (setting ǫ = 0) the entropy production ness, cosmological constant, and entropy problems, he
during the transition is s+ /s− = (1 + ǫ− Λ)
3/4
. In the pe- does go considerably further than we have in terms of
−
riod between t = tP and the transition, ǫΛ increases like specific model building. Moffat’s model does not satisfy
a4 , and the entropy density is diluted like 1/a3 . Hence our prescription for solving the cosmological problems,
after the transition the entropy density is what it was at but it may do so in a modified form. We are currently
+ investigating this possibility. We regret that because we
t = t− −3
P , that is s ≈ 1/LP . If we now follow the Uni-
verse until its Hubble time is t+ + were unaware of this work we did not cite it in the first
P (when its density is ρP )
we must wait until the expansion factor has increased by publicly distributed version of this paper.
a factor of (ρ+ − 1/4
P /ρP ) . Given that s ∝ 1/a3 the entropy
density is diluted by a factor of (ρ+ − 3/4
P /ρP ) . Therefore
ACKNOWLEDGEMENTS
the entropy density when the Hubble time is t = t+ P is
s ≈ 1/L+3 P . Again the dimensionless entropy inside the
Hubble volume, when this has size L+ We would like to thank John Barrow, Ruth Dur-
P , is of order 1.
Finally it is worth noting that treating the pre- rer, Robert Brandenberger, Gary Gibbons, and Erick
transition universe simply as a Roberston-Walker model Weinberg for discussion. We acknowledge support from
is no doubt overly simplistic, and we use it simply as PPARC (A.A.) and the Royal Society (J.M.).
a device to introducing our ideas. We expect that fur-
ther development of these ideas could result in a radically
different view of the pre-transition phase (much as has
happened with the inflationary scenario). One interest-
ing observation is that one could avoid having multiple
Planck times by considering that G ∝ c4 . Such assump-

9
 APPENDIX I: A SPECIFIC REALIZATION OF δ (3) (x − x(τ )) euµ
jµ = (42)
VSL γ c

In this Appendix we set up a specific VSL theory. We This current3 is the current which couples to the gauge
first discuss the simple case of the electrodynamics of the field, and in the rest frame it equals e. Therefore it can-
point particle in Minkowski space time. We start from not be conserved, and indeed we have that
Bekenstein’s theory of variable α, and show how a VSL jµ
alternative could be set up. We highlight the subtleties ∂µ j µ = ∂µ e (43)
encountered in the VSL formulation. We then perform e2
the same exercise with the Einstein-Hilbert action. We Let us now postulate instead that a changing α is to
briefly consider the dynamics of the field ψ = c4 . Finally be interpreted as c ∝ h̄ ∝ α−1/2 , and that e and m are
we cast the key elements of our construction into a body to be seen as constants. Minimal coupling, in the above
of axioms. sense, would then prompt us to consider the action (36),
but with c = c(xµ ) everywhere, and e and m constants.
This action leads to equations:
A. Electrodynamics in flat space time
1 e
mẍµ = (mc2 ),µ + uν Fµν (44)
A changing α theory was proposed by Bekenstein [23] 2 c
based on the postulate of Lorentz invariance. The elec- with the electromagnetic tensor defined as
trodynamics of a point particle was first analyzed. If
Lorentz invariance is to be preserved then the particle Fµν = c(∂µ (Aν /c) − ∂ν (Aµ /c)) (45)
mass m and its charge e must be variable. In order to
preserve “minimal coupling” (reduction to standard elec- However the above construction is not complete. In
tromagnetism when α = const) one chooses the world line spite of the appearance of Eqns. (36), (44), and (45),
action Lorentz invariance is broken. This boils down to the
e fact that, say ∂µ is no long a 4-vector. Even if c were
L = −mc −uµ uµ + uµ Aµ
p
(36) to be regarded as a scalar, ∂µ would contain c in its
c
zero component, but not in its spatial components. The
with uµ = ẋµ , gµν = ηµν , e = e(xµ ), and m = m(xµ ). usual contractions leading to S could still be taken but
Minimal coupling means simply to take the standard ac- S would no longer be a scalar. This manifests itself in
tion and replace e and m by variables without breaking the equations (44) in the fact that in ẍµ there are terms
Lorentz invariance. e and m must then be scalar func- in ∂c which break Lorentz invariance.
tions. This action leads to equation: Since the action is not Lorentz invariant, a minimal
e coupling prescription cannot possibly be true in every
(mẋµ )˙ = −m,µ c2 + uν Fµν (37) coordinate system. Minimal coupling is now the state-
c ment that there is a preferred reference frame in which
with the electromagnetic field tensor defined as the action is to be obtained from the standard action
simply by replacing c with a field. Let us call this frame
1 the “light frame”. In regions in which c changes very lit-
Fµν = (∂µ (eAν ) − ∂ν (eAµ )) (38)
e tle changes in the action upon Lorentz transformations
are negligible. Hence all boosts performed upon the light
The electromagnetic action can therefore be defined as: frame become nearly equivalent and Lorentz invariance
−1
Z is recovered.
SEM = d4 xFµν F µν (39) The Maxwell equations in a VSL theory become
16π
1
Also, the particle action (36) may be written as a La- ∂µ (cF µν ) = 4πj µ (46)
grangian density: c
in the light frame. Given that Lorentz invariance is bro-
δ (3) (x − x(τ ))
Z
SM = d4 x (−mc2 + (e/c)uµ Aµ ) (40) ken, one can no longer expect the general expression for a
γ conserved current to take the form ∂µ j µ = 0. Indeed one
in which γ is the Lorentz factor. Maxwell’s equations are
then:
3
e∂µ (F µν /e) = 4πj µ (41) There is an alternative view in which rather than a chang-
ing e one considers that the vacuum is a dielectric medium
with the current with variable ǫ. One may then identify a conserved charge,
but this is not the charge which couples to the gauge field.

10
could try and compute ∂ν of equations (46), but now ∂µ
and ∂ν do not commute. Also their commutator is not
Lorentz invariant: for instance [∂0 , ∂i ] = (−∂i c/c2 )∂0 .
Still, ∂µ j µ = 0 holds in the time frame. It is just that
this expression transforms into something more compli-
cated in other frames. The more complicated expression
would still place constraints on the theory, which could
still be called “conservation of charge”.
B. Minimal coupling to gravity
Let us now examine gravity in such a theory4 . As
in the previous case we will impose a minimal coupling
principle. Working in analogy with Brans-Dicke theory,
let us define a field ψ = c4 , and introduce the following
action
4 √
   
ψ(R + 2Λ)
Z
S = dx −g + LM + Lψ (47)
16πG
The dynamical variables are a metric gµν , any matter
field variables contained in LM , and ψ itself. The Rie-
mann tensor (and the Ricci scalar) is to be computed
from gµν at constant ψ in the usual way.
As in the previous section covariance is broken, in spite
of all appearances. ψ does not appear in coordinate
transformations of the metric, and so the connection Γα µν
does not contain terms in ∇ψ in any frame. However the
connection will contain different terms in ψ in different
frames. Hence the statement that the Riemann tensor
is to be computed from the metric at constant ψ can
only be true in one preferred frame. Minimal coupling
requires the definition of a light frame. The action (47)
is only Lorentz invariant in appearance.
Varying the action with respect to the metric leads to:
√ developped in our paper.
δS −gψ
= [Gµν − gµν Λ] (48)
δg µν 8πG
√ K = Λ = 0 one has
δSM −gψ
= − Tµν (49)
δg µν 8πG
leading to a set of Einstein’s equations without any extra
terms
8πG
Gµν − gµν Λ = Tµν (50)
ψ are solved, in the radiation epoch, c and h̄ should be
valid in the light frame. This is the way we chose to
phrase our postulates in SectionIII of our paper. In other
4 During phase transitions the perfect fluid approxima-
Gravitation is normally regarded as the gauge theory of
the Poincare group [24]. Here we simply abandon this point
of view. In some future work we will try to define a gauge
principle for broken symmetries, thereby recovering the stan-
dard view
11
words all we need is minimal coupling at the level of
Einstein’s equations.
The fact that a favoured set of coordinates is picked
by our action principle is not surprising as Lorentz in-
variance is broken. On the other hand notice that the
dielectric vacuum of Bekenstein theory is an ether the-
ory. His theory also breaks Lorentz invariance, not at
the level of the laws of physics, but in the form of the
contents of space-time.
In changing α theories a favoured frame is always
picked up. In a cosmological setting it makes sense to
identify this frame with the cosmological frame. Free
falling observers comoving with the cosmological flow de-
fine a proper time and a set of spatial coordinates, to be
identified with the light frame. In this frame the Einstein-
Hilbert action is minimally coupled to a changing ψ, and
the same happens to Friedmann equations. The rest of
our paper follows.
C. The dynamics of ψ
The definition of Lψ controls the dynamics of ψ. This
is the most speculative aspect of our theory, but it also
opens the doors to empirical model building. In our
paper we preferred a scenario in which c changes in
an abrupt phase transition, but one could also imagine
c ∝ an . The latter scenario would result from a Brans
Dicke type of Lagrangian
−ω
Lψ = ψ̇ 2 (51)
16πGψ
(where ω is a dimensionless coupling) and is being inves-
tigated. Addition of a temperature dependent potential
V (ψ) would induce a phase transition, as in the scenario
However here we only make the following remarks,
which are independent of any concrete choice of Lψ . If
√
δLψ −gT
= (52)
δψ 4ψ
and so in the radiation dominated epoch (T = 0), once
K = Λ = 0, one should not expect driving terms for
the ψ equation. Hence once the cosmological problems
constants. Incidently, once the matter dominated epoch
(T 6= 0) is reached, ψ should perhaps start changing
again, with interesting observations consequences [22].
We are studying the phase space portraits of these cos-
mologies, when say Λ 6= 0, and with various Lψ .
tion must break down. One should then use, say, scalar
field theory (let’s call it φ). Now notice that terms in φ̇
will act as a source to ψ (as they contain the speed of
light). Hence whenever there is a phase transition and
the VEV of a field changes a large amount, one may
expect a large change in the speed of light, with most APPENDIX II: SCALAR PERTURBATION
choices of Lψ . A changing ψ associated with SSB could EQUATIONS FOR VSL MODELS
then solve the quantum version of the cosmological con-
stant problem, but this might require a rather contorted In this Appendix we derive the scalar cosmological
choice of Lψ . perturbation equations in VSL scenarios. We assume
K = Λ = 0, and use a gauge where the perturbed metric
is written as
D. Axiomatic formulation of VSL theories
ds2 = a2 [−(1 + 2AY )dη 2 − 2BY ki dxi dη + δij dxi dxj ]
Postulate 1. A changing α is to be interpreted as (53)
a changing c and h̄ in the ratios c ∝ h̄ ∝ α−1/2 . The
coupling e is constant. for a Fourier component with wave vector k i . Here Y is a
This postulate merely sets up the theoretical interpre- scalar harmonic. We shall use conformal time η to study
tation of the possible experimental fact that α changes, in fluctuations, and denote ′ = d/dη. The stress energy
terms of variable dimensional quantities. This is a matter tensor is also writen as
of convention and not experiment, as much as a constant
h̄c is a matter of convention. With the above choice a δT00 = −ρY δ
system of units for mass, length, time, and temperature
 p v i
δT0i = − ρ + 2 kY
is unambiguously defined. c c
Postulate 2. There is a preferred frame for the laws δTji = pΠL Y δji + (k i kj − 1/3δji k 2 )Y pΠT (54)
of physics. This preferred frame is normally suggested by
the symmetries of the problem, or by a criterium such as The Einstein’s constraint equations then read [13]
c = c(t).  2
If c is variable, Lorentz invariance must be broken. 3 a′ 1 a′ 4πGa2
A − kB = − ρδ (55)
Even if one writes Lorentz invariant looking expressions c2 a c a c2
these do not transform covariantly. [In general this boils  ′ ′  ′ 2 !
k a′ a a B 4πGa2  p v
down to the explicit presence of c in the operator ∂µ . A− − = ρ + (56)
Once one admits that Lorentz invariance must be explic- c a a a c2 c2 c2 c
itly broken then a preferred frame must exist to formu-
late the laws of physics. These laws are not invariant and the dynamical equations are
under frame transformation, and one may expect that a
8πGa2 p ΠT
 
1 a′
preferred frame exists where these laws simplify. A+ B′ + 2 B = − (57)
Postulate 3. In the preferred frame one may obtain kc a c2 c2 k 2
 ′ 2 !
the laws of physics simply by replacing c in the standard a′ A′ a′′ a A 4πGa2 p
(Lorentz invariant) action, wherever it occurs, by a field 2
+ 2 − 2
= (ΠL − 2ΠT ) (58)
a c a a c c2 c2
c = c(xµ ).
This is the principle of minimal coupling. Because the We assume that these equations do not receive correc-
laws of physics cannot be Lorentz invariant it will not tions in ċ/c. This statement is gauge-dependent, much
hold in every frame. Hence the application of this postu- like its counterpart for the unperturbed Eintein’s equa-
late depends crucially on the previous postulate supply- tions. We can only hope that the physical result does
ing us with a favoured frame. This principle may apply in not change qualitatively from gauge to gauge. Comply-
Minkowski space time electrodynamics, scalar field the- ing with tradition we now define the comoving density
ory, etc, in which case the frame in which c = c(t) is contrast [13]
probably the best choice. The cosmological frame, en-
dowed with the cosmic proper time is probably the best a′ 1  v 
choice in a cosmological setting. ∆ = δ + 3(1 + w) −B (59)
ca k c
Postulate 4 The dynamics of c must be determined by
an action principle deriving from adding an extra term We also introduce the entropy production rate
to the Lagrangian which is a function of c only.
c2s
This is work in progress. We do not wish to specify Γ = ΠL − δ (60)
this postulate further because for all we know this extra wc2
term can be anything. We merely specify that no fields where the speed of sound cs is given by
(including the metric) must be present in this extra term  
because we wish minimal coupling to propagate into the 2 p′ 2 2 1 c′ a
cs = ′ = wc 1 − (61)
Einstein’s equations. ρ 3 1 + w c a′
Note that the thermodynamical speed of sound is given
by c2s = (∂p/∂ρ)|S . Since in SBB models evolution is

12
isentropic c2s = (∂p/∂ρ)|S = ṗ/ρ̇ = p′ /ρ′ . When ċ 6= 0 We assume that these do not receive ċ/c corrections. The
evolution need not be isentropic. However we keep the conservation equation is then:
definition c2s = p′ /ρ′ since this is the definition used in
perturbative calculations. One must however remember a′ c′ kc w
v ′ + (1 − 3w) v − 2 v = − ΠT (73)
that the speed of sound given in (61) is not the usual a c 2 1+w
thermodynamical quantity. With this definition one has
for adiabatic perturbations δp/δρ = p′ /ρ′ , that is the
ratio between pressure and density fluctuations mimics
the ratio of their background rates of change.
Combining all four Einstein’s equations we can then
obtain the (non-)conservation equations [13]

a′ c′

a′ [1] P. Steinhardt, Cosmology at the Crossroads, in the Pro-
′
∆ − 3w + ∆ = −(1 + w)kv − 2 wΠT (62) ceedings of the Snowmass Workshop on Particle As-
a c a
trophysics and Cosmology, E. Kolb and R.Peccei, eds.
c2s k
 ′    
′ a c′ 3 a′ a′ c′ (1995).
v + −2 v= − + ∆
a c 1 + w 2k a a c [2] A. Guth, Inflationary Universe: a possible solution to the
 ′ 2 ! horizon and flatness problem, Phys.Rev. D23 347 (1981).
kc2 w 2/3 3 a
+ Γ − kc + 2 2 wΠT (63) [3] A. Linde, A new inflationary scenario: a possible solu-
1+w 1+w k c a tion of the horizon, flatness, homogeneity, isotropy, and
primordial monopole problems, Phys. Lett B 108, 1220
These can then be combined into a second order equation (1982).
for ∆. If Γ = ΠT = 0 this equation takes the form [4] A. Albrecht and P. Steinhardt, Cosmology for Grand
Unified theories with radiatively induced symmetry
∆′′ + f ∆′ + (g + h + c2s k 2 )∆ = 0 (64) breaking, Phys.Rev.Lett. 48 1220 (1982).
[5] A. Linde, Chaotic inflation, Phys. Lett B 129, 177
with (1983).
[6] Of course if c− = ∞ we could in fact see the whole Uni-
c′ verse before tc . This also happens for a deSitter initial
f = (1 − 3w) − 3 (65)
c phase, where the event horizon diverges at the BigBang.
 ′ 2 
[7] C.M.Will, Theory and experiment in gravitational

a 9 2 3
g= w − 3w − (66) physics, CUP, Cambridge, 1991.
a 2 2
 ′ 2   ′ ′  ′ ′ [8] V. Mukhanov and R. Brandenberger, A non-singular
c 9w 5 c a c Universe, Phys. Rev. Lett. 68 1969 (1992).
h=2 + − − (67)
c 2 2 c a c [9] J. Levin and K. Freese, Curvature and flatness in Brans-
Dicke Universe, Nucl. Phys. B421 635 (1994). J. Levin
Phys. Rev D51 462 (1995). J. Levin Phys. Rev D51 1536
(1995).
APPENDIX III: VECTOR PERTURBATION [10] Y. Hu, M.S. Turner, and E. Weinberg, Dynamical solu-
EQUATIONS FOR VSL MODELS tions to the horizon and flatness problems, Phys. Rev.
D49 3830 (1994).
In a similar fashion we can study vector modes in a [11] S. Weinberg, Theories of the cosmological constant, in
Critical dialogues in cosmology, ed. Neil Turok, PUP
gauge where the metric may be written as
Princeton.
ds2 = a2 [−dη 2 + 2BYi dxi dη + δij dxi dxj ] (68) [12] V. Mukhanov and G. Chibisov, Zh. Eksp. Teor. Fiz 83
(1982) 475. S. W. Hawking, Phys. Lett B 115 (1982) 295.
where Yi is a vector harmonic. The stress energy tensor A. Starobinsky, Phys. Lett. B 117 (1982) 175. A. Guth
is written as and S.-Y. Pi Phys. Rev. Lett. 49, (1982) 1110. J. Bardeen,
P. Steinhardt, and M. Turner, Phys. Rev. D 28 (1983)
 p  v 
679.
δTi0 = ρ + 2 − B Yi (69)
c c [13] H.Kodama and M.Sasaki, Prog.Th.Phys. Supp., 78 1-166
δTij = pΠT Y(i,j) (70) (1984).
[14] J. Peebles, Principles of Physical Cosmology, 371-373,
Einstein’s equations then read [13] eds, PUP, Princeton.
[15] A. Vilenkin and E.P.S. Shellard, Topological defects in
16πG 2  p v Cosmology, CUP, (1996).
k2 B = a ρ + (71) [16] A. Albrecht in the proceedings of the ”International
c2 c2 c
k

a ′

8πG p Workshop on Particle Physics and the Early Universe”
B ′ + 2 B = − 2 2 a2 ΠT (72) Ambleside, September 1997, L. Roszkowski Ed. (1998)
c a c c astro-ph/9802135

13
[17] J. Barrow, Why the Universe is not anisotropic,
Phys.Rev. D 51 (1995) 3113.
[18] W.A.Baum and R.Florentin-Nielsen, Astroph.J. 209, 319
(1976).
[19] J.E.Solheim, T.G.Barnes III, and H.J. Smith, Astro-
phys.J. 209 330 (1976).
[20] J.D.Bekenstein, Com. on Astroph. VIII, 89 (1979).
[21] R.H.Dicke, Phys.Rev 125, 2163 (1962).
[22] J.K.Webb et al., Evidence for Time Variation of the Fine
Structure Constant, astro-ph/9803165.
[23] J.D.Bekenstein, Phys.Rev. D 25, 1527 (1982).
[24] T.Kibble, J.Math.Phys. 2 212 (1961).
[25] J.Barrow,Cosmologies with varying light speed, submit-
ted to Physical Review D, 1998.
[26] J. D. Barrow and J. Magueijo,“Varying-α theories and
solutions to the cosmological problems”, Phys. Lett. B,
IN PRESS.
[27] J. Moffat, International Journal of Physics D, Vol. 2, No.
3 (1993) 351-365; Foundations of Physics, Vol. 23 (1993)
411.

14